5 Feb 2024 14:00-15:00 @ COM3 02-59: Doug Oard / Balancing Information Protection and Information Access

[Min Yen KAN]

Hi all:

Just to let you all know that Doug Oard will pass by Singapore in early Feb and will visit SoC on 5 Feb. Some of you might know him. Prof. Chua Tat-Seng has arranged the seminar from 2:00-3:00pm.  (This message passed on to us via Tat-Seng)

– Min

Title: Balancing Information Protection and Information Access
Speaker: Prof. Douglas W. Oard, University of Maryland
Date/Time: 5 Feb, 2024, Monday, 2:00 PM to 3:30 PM
Venue: COM3-02-59 - Meeting Rm 20 @ COM3

Chaired by: Prof. Tat-Seng Chua, School of Computing

Abstract:

Current search engines are designed to find things, but there are many cases in which we actually don’t want some things to be found.  In particular, we can’t yet make many potentially valuable collections available to be searched because they contain some intermixed sensitive content that requires protection, but for which reliable sensitivity labels are not available. Some prominent examples include government transparency regimes such as the Freedom of Information Act in the United States, the rapidly growing backlog of national security information in the United States that is awaiting declassification review, and the vast troves of email that are now accumulating in both government archives and personal collections.  The scale of many of these problems is such that asking people to mark all the sensitive content in a collection would simply be impractical.  If we are ever to be able to find that which is not actually sensitive, we will thus need to build systems—systems involving both people and automation—that are able to recognize and protect that which requires protection.  We formulate this as a multi-objective optimization problem in which the goal is to balance information access with information protection.  I’ll describe two implementations of this broad idea.  In the first, designed for high-stakes tasks such as topic-focused declassification review, the search is performed on behalf of the end user by a trusted intermediary (e.g., an archivist), and the system’s goal is to focus that intermediary’s limited time in a way that balances the risk of missing relevant content with the risk of revealing sensitive content.  In the second, designed for high-volume but lower-stakes cases such as searching archived email, we seek to support end-user search by using a risk-averse search engine to surface some relevant content, thus allowing searchers to explore collections to find some immediately useful content, with the beneficial side effect that their final refined queries might also be used to flag the more difficult decisions for (future) human review. Both of these techniques require automated sensitivity classifiers, so I will also describe two lines of work on that problem: one in which sensitivity is a property of the content itself, and a second in which it is not the content itself that is sensitive, but rather the inferences that could be drawn if that content were to be released.  This is joint work with Jason Baron, Mahmoud Sayed, Nate Rollings, Fabrizio Sebastiani and Jyothi Vinjumur.

Biodata:

Doug Oard is a Professor at the University of Maryland, with joint appointments in the College of Information Studies and the University of Maryland Institute for Advanced Computer Studies (UMIACS).  He earned his Ph.D. in Electrical Engineering from the University of Maryland, and his research interests center around the use of language technologies such as speech recognition, machine translation, document image analysis, knowledge representation, processing mathematical notation, and social network analysis to support information access to support information seeking by end users. More on Doug’s research can be found at http://terpconnect.umd.edu/~oard.